JAMA Network adds a dedicated venue for AI evidence in medicine
The JAMA Network is carving out a dedicated home for a fast-expanding corner of medicine. On Oct. 15, the network announced the launch of JAMA AI, a peer-reviewed journal focused on artificial intelligence and emerging healthcare technologies, with manuscript submissions set to open in early 2027. The journal will be online-only and will become the 14th publication in the JAMA Network.
The timing reflects a real tension in modern care delivery. AI tools are already moving into clinical workflows, yet the evidence base for safety, effectiveness and downstream impact often lags behind the pace of adoption. In practice, that gap leaves clinicians, health systems and patients trying to judge tools that may influence diagnosis, treatment selection, triage, documentation and resource allocation without the kind of rigorous journal home that has historically shaped standards in other areas of medicine.
JAMA AI is being designed to fill that space with research that is explicitly method-focused as well as application-focused. The journal will publish work on generative AI, agentic AI, clinical decision support, AI-enabled diagnostics and therapeutics, and digital health technologies. It will also cover methods used to evaluate safety and effectiveness, along with the effects these tools have on patients, clinicians and health systems. That scope matters because AI in medicine is not a single intervention; it spans software that drafts text, systems that propose actions, and tools that influence clinical judgment at the bedside.
The move also signals an editorial recognition that the central question is no longer whether AI will enter healthcare. It already has. The sharper question is how to build an evidence standard strong enough to distinguish useful tools from those that are merely novel, efficient or easy to deploy. In a field where model outputs can change over time, where performance may drift across settings, and where clinical context can determine whether a recommendation helps or harms, publication standards need to track both technical performance and patient-facing consequences.
Roy Perlis, MD, director of the Center for Quantitative Health at Massachusetts General Hospital in Boston, will serve as the inaugural editor in chief. On Feb. 1, 2027, he will become director of the Bakar Computational Health Sciences Institute at the University of California San Francisco. His appointment places the journal under leadership with direct ties to quantitative health and computational medicine, two areas that sit close to the practical challenges of evaluating AI in real-world care.
Why an AI journal now carries clinical weight
Medicine has seen many technologies arrive with speed and confidence, only to require years of careful study before they could be used safely at scale. AI is following a similar pattern, but with a much broader reach. A single model can affect documentation, imaging interpretation, triage, medication decisions or patient communication, which means failures can propagate quickly through systems if evaluation is thin.
A peer-reviewed journal dedicated to this space creates a formal path for studies that may not fit neatly into conventional clinical silos. Work on algorithmic performance, workflow integration, human factors, fairness, monitoring, implementation science and outcome measurement all need a venue that treats them as clinically relevant rather than merely technical. By naming safety and effectiveness as central publishing priorities, JAMA AI is also signaling that flashy demonstration projects are not enough; the field will need evidence that withstands scrutiny across settings and populations.
The launch arrives at a moment when the demand for rigorous evidence has become increasingly urgent as AI adoption accelerates across medicine. That urgency is easy to understand from a bedside perspective. Clinicians do not just need tools that perform well in a test environment. They need systems that behave predictably under pressure, align with clinical workflow, and avoid introducing new forms of error, bias or overreliance. When tools are aimed at supporting rather than replacing clinicians, the quality of the supporting evidence becomes the difference between augmentation and confusion.
The journal’s focus on methods for evaluating AI is especially important. In healthcare, evaluation cannot stop at accuracy metrics alone. A diagnostic algorithm may appear strong in retrospective analysis but still fail when deployed in a different hospital, among different patient groups or in a workflow with different documentation habits. A therapeutic decision-support system may improve efficiency while worsening inappropriate variation in care. A digital health tool may be widely adopted but deliver uncertain clinical benefit. Those problems require publication venues that prize measurement discipline as much as novelty.
Dr. Perlis also serves as the founding editor in chief of JAMA+ AI, the channel the network launched in 2024 to highlight AI research across its journals. That channel will continue operating, giving the network a dual structure: one broad channel for ongoing coverage and one dedicated journal for full peer-reviewed research. The arrangement suggests a pipeline that can surface timely work while also reserving space for deeper, more formal studies of emerging technologies.
What the launch says about standards for digital medicine
JAMA AI is entering a crowded technological moment, but the editorial challenge is distinct. Many technology announcements emphasize capability. Clinical publishing has to ask whether capability translates into benefit, and whether benefit persists when the tool is used by different clinicians, in different settings, and under different clinical pressures.
The journal’s planned emphasis on patients, clinicians and health systems broadens the frame beyond software performance alone. AI can alter workload, decision timing, communication patterns and operational flow. It can change how quickly a clinician reaches an answer, how often a recommendation is accepted, and how much trust a team places in a machine-generated output. Those effects are not side issues; they are part of the clinical consequence of using the technology.
By opening submissions in early 2027, the JAMA Network is giving the field time to produce research that can meet a stricter evidentiary standard. The first wave of AI papers in medicine has already shown how rapidly interest can outpace proof. A dedicated peer-reviewed journal will not solve that problem on its own, but it gives investigators a place to submit studies built around safety, effectiveness and real-world impact rather than speculation.
The remaining challenge is practical and immediate: the field still needs durable evidence on which AI systems improve care, which ones create new risks, and which outcomes matter most when these tools move from pilot projects into routine use. JAMA AI is being created to publish exactly that kind of work, and its early output will show whether the journal becomes a true evidentiary filter for digital medicine or simply another outlet for the same uneven science.